ED 036 522

AUTHORTITLE

INSTITUTION

SPONS AGENCY

REPORT NOBUREAU NOPUB DATECONTRACTNOTE

EDES PRICEDESCRIPTORS

ABSTRACT

DOCUMENT RESUME

24 TE 001 728

BLCUNT, NATHAN S.; AND OTHERSMEASURES OF WRITING MATURITY FROM TWO 500-WORDWRITING SAMPLES. REPORT FROM THE INDIVIDUALLY GUIDEDINSTRUCTION IN ENGLISH LANGUAGE, COMPOSITION, ANDLITERATURE PROJECT.WISCONSIN UNIV., MADISON. RESEARCH AND DEVELOPMENTCENTER FOR CCGNITIVE LEARNING.OFFICE OF EDUCATION (DHEW) , WASHINGTON, D.C. BUREAUOF RESEARCH.TR-S7BR-5-0216SEP 69CEC-5-10-15414P.

EDES PRICE MF-$0.25 EC-$0.80COMECSITION (LITERARY), *COMPOSITION SKILLS(LITERARY) , EVALUATICN CRITERIA, *EVALUATIONMETHODS, GRADE 8, MATURITY TESTS, *MEASUREMENTTECHNIQUES, *RELIABILITY, *SAMI,LING, SENTENCESTRUCTURE

THE PURPOSE OF THIS STUDY WAS TO ESTIMATE THERELIABILITY OF VARIOUS MEASURES CF WRITING BEHAVIOR USING 500-WORDSAMPLES INSTEAD OF 1000-WORE SAMPLES. THE THEMES OF 135 EIGHTHGRADERS ON NARRATIVE AND EXPOSITC11Y TOPICS WERE COLLECTED FOR 6WEEKS, ONE THEME PER WEEK., SAMPLES CF AT LEAST 1000 WORDS WEREOBTAINED FROM EACH STUDENT, AND gHE CRITERIA OF HUNT (1965) WERE USEDTO DETERMINE WHICH SENTENCES AND T-UNITS TO INCLUDE FOE TABULATION.TWO 500WORD DIVISIONS OF THE 1000-WORD SAMPLES WEEE MADE AND TWOCOMPARISONS BETWEEN THE DIVISICNS WERE CARRIED CUT. THE FIRSTCOMPARISCN INVOLVED SPLITTING THE SAMPLE AT THE SENTENCE OR T-UNITNEAREST THE 500-WORD MIDPCINT. THE SECOND COMPARISON WAS BETWEEN THECDD PAGES ANE THE EVEN PAGES. ADJECTIVE, ADVERB, AND NOUN CLAUSESOCCURRED FREQUENTLY ENOUGH TO BE OF IMPORTANCE AND CORRELATED .35,.31, AND .08, RESPECTIVELY, IN THE FIRST-HALF/SECOND-HALF SAMPLES,AND .46, .42, AND .26 IN THE ODD/EVEN PAGE SAMPLES. THE FIVE SYNOPSISSCORES (CLAUSE LENGTH, CLAUSES PER T-UNIT, T-UNIT LENGTH, T-UNITS PERSENTENCE, ANE SENTENCE LENGTH) CORRELATED .48, .22, .56, .48, AND 62BETWEEN FIRST -HALF AND SECOND -HALF RESPECTIVELY, ANL .69, .54, .74,.65, AND .77 BETWEEN ODD-PAGE/EVEN-PAGE SAMPLES. (THREE TABLES AREINCLUDED.) (AUTHOR/LH)

U.S. DEPARTMENT OF HEALTH, EDUCATION 31 WELFARE

OFFICE OF EDUCATION

THIS DOCUMENT HAS BEEN REPRODUCED EXACTLY AS RECEIVED FROM THE

PERSON OR ORGANIZATION ORIGINATING IT. POINTS OF VIEW OR OPINIONS

STATED DO NOT NECESSARILY REPRESENT OFFICIAL OFFICE OF EDUCATION

POSITION OR POLICY.

Technical Report No. 97

MEASURES OF WRITING MATURITY FROM

TWO 500WORD WRITING SAMPLES

b

S- 6 ;L/6-74 97

e-///3/2_

Nathan S. Blount, Wayne C. Fredrick, & Shelby L. Johnson

Report from the Individually Guided Instruction inEnglish Language, Composition, and Literature Project

Nathan S. Blount and Lester S. Golub, Principal Investigators

Wisconsin Research and DevelopmentCenter for Cognitive LearningThe University of Wisconsin

Madison, Wisconsin

September 1969

Published by the Wisconsin Research and Development Center for Cognitive Learning, supportedin part as a research and development center by funds from the United States Office of Education,Department of Health, Education, and Welfare. The opinions expressed herein do not necessarilyreflect the position or policy of the Office of Education and no official endorsement by the Officeof Education should be inferred.

Center No. C-03 / Contract OE 5-10-154

NATIONAL EVALUATION COMMITTEESamuel BrownellProfessor of Urban EducationGraduate SchoolYale University

Launor F. CarterSenior Vice President on

Technology and DevelopmentSystem Development Corporation

Francis S. ChaseProfessorDepartment of EducationUniversity of Chicago

Henry ChaunceyPresidentEducational Testing Service

Martin DeutschDirector, Institute for

Developmental StudiesNew York Medical College

Jack EdlingDirector, Teaching Research

DivisionOregon State System of Higher

Education

Elizabeth KoontzPresidentNational Education Association

Roderick McPheePresidentPunahou School, Honolulu

G. Wesley SowardsDirector, Elementary EducationFlorida State University

Patrick SuppesProfessorDepartment of MathematicsStanford University

*Benton J. UnderwoodProfessorDepartment of PsychologyNorthwestern University

UNIVERSITY POLICY REVIEW BOARD

Leonard Berkowitz John Guy Fowlkes Herbert J. Klausmeier M. Crawford YoungChairman Director Director, R & D Center Associate DeanDepartment of Psychology Wisconsin Improvement Program Professor of Educational The Graduate School

Psychology

Archie A. Buchmiller Robert E. Grinder Donald J. McCartyDeputy State SuperintendentDepartment of Public Instruction

*James W. ClearyVice Chancellor for Academic

Affairs

Leon D. EpsteinDeanCollege of Letters and Science

ChairmanDepartment of Educational

Psychology

H. Clifton HutchinsChairmanDepartment of Curriculum and

Instruction

Clausten JenkinsAssistant DirectorCoordinating Committee for

Higher Education

DeanSchool of Education

Ira SharkanskyAssociate Professor of Political

Science

Henry C. WeinlickExecutive SecretaryWisconsin Education Association

EXECUTIVE COMMITTEE

Edgar F. BorgattaBrittingham Professor of

Sociology

Max R. GoodsonProfessor of Educational Policy

Studies

Russell J. HosierProfessor of Curriculum and

Instruction rnd of Business

*Herbert J. KlausmeierDirector, R & D CenterProfessor of Educational

Psychology

FACULTY OF PRINCIPAL INVESTIGATORS

Ronald R. AllenAssociate Professor of Speech

and of Curriculum andInstruction

Vernon L. AllenAssociate Professor of Psychology(On leave 1968.69)

Nathan S. BlountAssociate Professor of English

and of Curriculum andInstruction

Robert C. CalfeeAssociate Professor of Psychology

Rober E. DavidsonAssistant Professor of

Educational Psychology

Gary A. DavisAssociate Professor of

Educational Psychology

M. Vera DeVaultProfessor of Curriculum and

Instruction (Mathematics)

Frank H. FarleyAssistant Professor of

Educational Psychology

John Guy Fowlkes (Advisor)Professor of Educational

AdministrationDirector of the Wisconsin

Improvement Program

Lester S. GolubLecturer in Curriculum and

Instruction and in English

Wayne OttoProfessor of Curriculum and

Instruction (Reading)

Robert G. PetzoldAssociate Dean of the School

of EducationProfessor of Curriculum and

instruction and of Musk

Max R. GoodsonProfessor of Educational Policy

Studies

Warren 0. HagstromProfessor of Sociology

John G. HarveyAssociate Professor of

Mathematics and Curriculumand Instruction

Herbert J. KlausmeierDirector, R & D Center

Professor of EducationalPsychology

Burton W. KreitlowProfessor of Educational Policy

Studies and of Agriculturaland Extension Education

Richard L. VenezkyAssistant Professor of English

and of Computer Sciences

Richard G. MorrowAssistant Professor of

Educational Administration

Wayne OttoProfessor of Curriculum and

Instruction (Reading)

Milton 0. PellaProfessor of Curriculum and

Instruction (Science)

Thomas A. RombergAssistant Professor of

Mathematics and ofCurriculum and Instruction

Richard L. ItenezkyAssistant Professor of English

and of Computer Sciences

MANAGEMENT COUNCIL*Herbert J. Klausmeier

Director, R & D CenterActing Director, Program 1

Mary R. QuillingDirectorTechnical Section

Thomas A. RombergDirectorPrograms 2 and 3

James E. WalterDirectorDissemination Section

Dan G. WoolpertDirectorOperations and Business

* COMMITTEE CHAIRMAN

STATEMENT OF FOCUS

The Wisconsin Research and Development Center for Cognitive Learningfocuses on contributing to a better understanding of cognitive learning bychildren and youth and to the improvement of related educational practices.The strategy for research and development is comprehensive. It includesbasic research to generate new knowledge about the conditions and processesof learning and about the processes of instruction, and the subsequent develop-ment of research-based instructional materials, many of which are designed foruse by teachers and others for use by students. These materials are tested andrefined in school settings. Throughout these operations behavioral scientists,curriculum experts, academic s.cholars, and school people interact, insuringthat the results of Center activities are based soundly on knowledge of subjectmatter and cognitive learning and that they are applied to the improvement ofeducational practice.

This Technical Report is from the Individually Guided Instruction in EnglishLanguage, Composition, and Literature Project in Program 2. General objectivesof the Program are to establish rationale and strategy for developing instructionalsystems, to identify sequences of concepts and cognitive skills, to developassessment procedures for those concepts and skills, to identify or L ,velop in-structional materials associated with the concepts and cognitive skills, and togenerate new knowledge about instructional procedures. Contributing to theseProgram objectives, the long-range objective of the English Project is to installand test materials for individually guided instruction in language, composition,and literature. Prerequisite activities include formulating behavioral objectivesfor students and teachers, based on a content and concepts outline, and devel-oping measurement instruments related to the behavioral objectives.

iii

CONTENTS

Page

List of Tables vii

Abstract ix

I The Problem1

II Experimental Procedure 2

III Results and Discussion 4

References 9

LIST OF TABLES

Table

1 Means and Standard Deviations Tabulated from the 500-WordWriting Samples of 135 Eighth Graders

2 Correlations for 17 Pairs of Writing Measures from Two 500 -Word Samples

3 Intercorrelations of the Five Synopsis Scores in the Odd-EvenComparison

Page

4

5

6

vii

ABSTRACT

The purpose of this study was to estimate the reliability of variousmeasures of writing behavior using 500-word samples instead of 1000 words.The themes of 135 Eighth Graders were collected over a period of 6 weeks,one theme per week. The topics, expository and narrative, were selectedjointly by teachers and students as part of the normal curriculum in English.Samples of at least 1000 words were obtained from each student and thecriteria of Hunt (1965) were\ used to determine which sentences and T-unitsto include for tabulation. Two 500-word divisions of the 1000-word sampleswere made. The themes written during the first 3 weeks were compared tothose of the second 3 weeks, and the odd pages were compared to the evenpages.

Eight of the tabulated structures (clauses of comparison, deferred sub-ject, cleft sentence, "the more, the merrier" construction, subjunctive wordorder, special which clauses, adjective complements, and adverb comple-ments) occurred very infrequently and were essentially uncorrelated (r < .26)from sample to sample. Adjective, adverb, and noun clauses were more fre-quent and correlated .35, .31, and .08, respectively, in the first-half/second-half samples, and .46, .42, and .26 in the odd/even pages samples. Thefive synopsis scores (clause length, clauses per T-unit, T-'snit length,T-units per sentence, and sentence length) correlated .48, .22, .56, .48,and .62 between first half and second half respectively, and .69, .54, .74,.65, and .77 between odd-page and even-page samples.

ix

THE PROBLEM

Hunt (1965) used 1000-word writing samplesfrom which he tabulated eleven syntacticalstructures and five synopsis scores. The workrequired to tabulate such a large amount ofwriting is arduous, and if a smaller sample canbe used which will still give an accurate meas-ure of writing behavior, much menial labor canbe spared. The purpose of the present studywas to provide some data on the reliability ofthe Hunt measures when 500 words of 1000 -word samples are used.

Some very old studies have made inferencesabout what length a writing sample should be.Frogner (1933} thought that for some measuresone needed to analyze 50,000 to 60,000 wordsbefore an accurate measure of group perform-ance was achieved, She made no estimate ofthe sample size required from an individual.This high number of words was especiallyneeded when tabulating errors and specificcategories of clauses. Other measures, suchas the proportions of simple, compound, andcomplex sentences, produced by groups seemedto stabilize after 10,000 words had been tabu-lated. Frogner also noted that the kind of writ-ing influenced the complexity of the sentences.She showed that expository writing produced ahigher percentage of complex sentences thannarrative, which in turn produced a higher pro-portion of complex sentences than those pro-duced in letter-writing.

Anderson (1937) tested the feasibility ofusing samples from individuals as small as 150words to tabulate sentence length. He con-cluded that 150-word samples were too smallto reliably measure even the most objectiveindices. He postulated that sentence lengthwould vary with the situation and the subjectmatter.

Chotlos (1944) made a systematic compari-son of the reliability of type-token ratios (num-ber of unique words divided by the number ofwords) obtained from 100-, 500-, 1000-, and1500-word samples written by individuals. Thetype-token ratios for two 100-word samples

correlated .38, while two 500-word samplescorrelated .67, 1000-word .81, and 1500 -word .83. Though these correlations showeda clear increase with sample size, there wassome confounding involved since, as samplesize increased, the number of words betweensamples decreased, For example, the 1500 -word samples were the first and second halfof a 3000-word corpus, but the 1000 wordswere the first and last thirds of this same cor-pus. Thus, 1000 words of writing had takenplace between the two 1000-word samples, butnone had occurred between the 1500-word sam-ples. Whatever effect writing has on writing,it became greater with the smaller samples,until with 100 words a massive 2800 words hadintervened between the two samples. The type-token ratio is also critically dependent uponthe number of words in a way that other meas-ures (sentence length, clause length, T-unitlength) are not. The types of words are accumu-lated rapidly as one begins tabulating. Astabulating continues the rate at which newtypes of words are added declines, but wordtokens are _Idded at a constant rate throughout.

Recent studies have settled upon 1000 wordsper S as a suitable body of writing for purposesof analysis (Hunt, 1965; Blount, Fredrick, &Johnson, 1968). One can ask, however,whether a smaller sample will result in meas-ures that approach the reliability of the samemeasures obtained from a larger sample Ifa 500-word sample can replace a 1000-wordsample with only a slight decrease in relia-bility, the savings of time and effort are ob-vious.

Taking under consideration the usual levelof variability in measures of writing, it wasdecided that whenever a variable accountedfor one-fourth of the variance in two succes-sive samples (r > .5) that variable was of somevalue as a measure of group performance.Whenever one-half of the variance wasidentifiable (r > .7) the measure was of valuefor assessing individual performance.

1

EXPERIMENTAL PROCEDURE

The as were Eighth Graders in a large publicjunior high school in Wisconsin. High and lowability Ss were drawn from the Eighth Gradepopulation (242 students in the classrooms ofthe participating teachers). Among these were72 females and 63 males. Seventy-two Ss hadabove average scores (119-145) on the Cali-fornia Test of Mental Maturity and 63 were be-low average (90-106).

During the first part of the 1966-67 schoolyear each S wrote one in-class theme per weekas part of the normal work in his English class.Ss did not know that the themes were being col-lected from the teachers, copied, and returnedby the experimenters. The topics of the themeswere the usual descriptive, narrative, and ex-pository assignments decided upon jointly bythe students and teachers. The successiveweekly themes were collected until each of the135 Ss had produced at least 1000 words oftext. Generally this was accomplished byabout the sixth theme.

From the copies the themes were typed triple-spaced in preparation for the tabulation of vari-ous measures of writing ability. The guidelinesdeveloped by Hunt (1965) were used in tabulat-ing 17 measures, which included the followingeleven subordinate structures, the total amountof subordination, and five synopsis scores:

1. Adjective clauses: clause structureswhich functioned as adjectives inmodifying a noun or pronoun.<That was the place where he losthis keys.>

2. Adverb clauses: clause structureswhich functioned as movable adverbs,<I heard noises when I woke up.>

3. Noun clauses: clause structures whichoccurred in the nominal positions ofsubject, direct object, object of prep-osition, and appositive.<That there are good teenagers isnever mentioned.>

4. Clauses of comparison: adverb-likeclauses which were not usually mov-able and which used comparativessuch as so that, more than, as longas, etc.<The clouds seemed so low that youcould catch one.>

5. Deferred subject: clause structureswhich were the "logical" subjects ofthe sentence but which followed the"grammatical" subject it.< It was good that such a thing hap-pened.>

6. Cleft sentence: clause structures(other than "logical" subjects) causedby it as the "grammatical" subject.< It's the fog that keeps the fumes fromescaping.>

7. "The more, the merrier": sentenceswhich involved a the more this, themore that statement.<The more he worried, the more hefelt sorry for himself.>

8. Subjunctive: use of different wordorder to express the subjunctive mood.< I would have been killed had it notbeen for him.>

9. Special which clauses: which clausesthat modified an entire sentence.<When we come in the morning we cango to the gym, which I think is nice.>

10. Adjective complement: clauses thatcomplemented an adjective and couldnot be moved.<She was sure that I had left home.>

11. Adverb complement: clauses that fol-lowed verbs like seem, appeared,feel, etc.<She looked as if she had seen a ghost.>

12. All subordinate clauses: the total of1 - 11 above.

For the synopsis scores, the number of words,clauses, T-units, and sentences were tabu-lated according to these guidelines: Wordswere all letter sequences that, regardless ofshelling, could be considered a unit of speech.Contractions were counted as two words, andproper nouns (Yellowstone National Park) werecounted as one word. Clauses were all mainand subordinate structures that contained asubject and a finite verb phrase. T-units weredefined as "one main clause plus the subordi-nate clauses attached to or embedded withinit [Hunt, 1965, p. 49)." Sentences were thewords written between a capital letter and someend punctuation. As an example of each ofthese terms, note The following sentence whichcontains three T-units marked by double slashes,a main clause within each T-unit, three subor-dinate clauses marked by a single slash and asymbol, and 42 words: <If I had a million dol-lars (adv) / I think / (N) that I would do a lotof things, // but I would put most of the moneyin the First National Bank // and I would starta job / (adv) so people wouldn't call n:e a lazyman. //> The tabulations of words, clauses,T-units, and sentences were then used to com-pute these synopsis scores:

13. Clause length: words per clause.14. Clauses per T-unit.15. T-unit length: words per T-unit.

16. T-units per sentence.17. Sentence length: words per sentence.

Under Hunt's guidelines all sentence frag-ments, garbles, direct discourse, and ques-tions were deleted from consideration. The17 measures were tabulated independently bytwo raters who had been trained previouslyand had used the tabulating system in previouswork for the experimenters, In past work thecorrelations between the two raters hat.: rangedfrom .84 to 1.00 on the various indices tabu-lated. All discrepancies between the two raterswere resolved by the experimenter (ST) who hadoriginally trained the raters, A detailed dis-cussion of each measure can be found in BlountFredrick, and Johnson (1968).

When typed sentence-by-sentence and triple-spaced, each S's writing sample became a docu-ment of about 12 pages. Two comparisons be-tween halves of the 1000-word samples werethen made. The first comparison involved split-ting the sample at the sentence or T-unit nearestthe 500-word midpoint. Each of the 17 scoresobtained from the first half was correlated withthe respective score from the second half. Thesecond comparison was between the odd pagesof the typed sample and the even pages. Eachof the 17 scores from the odd pages (to the endof the sentence or T-unit nearest 500 words)was correlated with the respective score fromthe even pages.

3

III

RESULTS AND DISCUSSION

The means and standard deviations from135 odd-page samples from the odd-even pagessplit are presented in Table 1. The equivalenttables for even-page samples and first-half/second-half samples are not presented sinceall four tables were practically identical. Themeans presented include eleven tabulations ofsubordinate structures, the total of all elevenkinds of subordinate structures, and five syn-opsis scores. In 500 words of Eighth-Gradewriting, an average of 22 subordinate struc-tures appeared; roughly 5 of these were adjec-tive clauses, 9 adverb clauses, and 7 noun

Table 1

clauses. Each clause averaged about eightwords in length, and appeared with a frequencyof three clauses per every two T-units. T-units themselves averaged 13 words long, andone-third of the time there was more than oneT-unit in a sentence. Sentence length averagedmore than 16 words. Each of these statisticsclosely approximated the results found in pre-vious studies (Hunt, 1965; Blount, et al.,1968). [Adjustments were made to frequencycounts so that 500- and 1000-word samplescould be compared.]

Means and Standard Deviations Tabulated fromthe 500-Word Writing Samples of 135 Eighth Graders

VariableAll 135 Ss 63 Males 72 Females 72 High IQ 63 Low IQMean sd Mean sd Mean sd Mean sd Mean sd

Adjective Clauses 5.3 2.9 5.4 3.1 5.2 2.7 5.5 3.0 5.1 2.8Adverb Clauses 8.8 4.2 8.3 4.2 9.3 4.2 8.4 4.2 9.3 4.3Noun Clauses 7.0 3.8 6.7 4.0 7.2 3.6 6.8 3.8 7.2 3.8Clauses of Comparison .4 .7 .5 .7 .3 .7 .4 .7 .4 .7Deferred Subject .1 .3 .0 .2 .1 .4 .1 ,3 .1 .4Cleft Sentence .1 .3 .1 .4 .1 .3 .1 .3 .1 .4"The More, the Merrier" .0 .1 .0 .2 .0 - .0 .2 .0 -Subjunctive .0 .2 .0 .3 .0 - .0 - .0 .3Special Which Clauses .1 .2 .1 .2 .0 .2 .1 .2 .0 .2Adjective Complement .1 .4 .1 .3 .2 .5 .1 .4 .1 .5Adverb Complement .1 .2 .0 .2 .1 .3 .1 .2 .0 .2

All Subordinate Clauses 22.0 6.2 21.3 6.6 22.6 5.9 21.6 6.6 22.5 5.7

Clause Length 8.2 1.2 8.4 1.3 8.1 1.1 8.3 1.3 8.2 1.1Clauses per T-unit 1.6 .2 1.6 .2 1.6 .2 1.6 .2 1.6 .2T-unit Length 13.1 2.7 13.2 2.7 13.0 2.7 13.1 2.9 13.1 2.5T-units per Sentence 1.3 .2 1.3 .2 1.2 .2 1.2 .2 1.3 .2Sentence Length 16.4 3.5 16.9 3.7 16.0 3.3 16.0 3.1 16.9 3.9

4

To compare the differences between odd/even page samples, t tests were computed onall 17 dependent measures. These tests werenot significant nor were there significant dif-ferences in comparisons of the first -half /second --half samples. There were also no significantdifferences when the mean scores of the 63males were compared to the means of the 72females and neither were any significant dif-ferences observed in the comparison of highand low IQ Ss.

The feasibility of using 500-word samplebto obtain measures of subordination and writingmaturity can be seen more clearly in the corre-lation coefficients presented in Table 2. ThePearson product-moment correlations betweenthe samples in the odd/even split for the 11subordinate structures were far belOw accept-able levels for individual and group measure-ment. The first-half/second-half correlationsfor these 11 measures were even smaller thanthe corresponding correlations from the odd/even split. Thus, a 500-word sample did not

give a reliable measure of the use of certainsubordinate structures. Even when consider-ing all subordinate structures, the correlationbetween two 500-word samples only reached.46. By using the Spearman-Brown prophecyformula, it was estimated that two 1000-wordsamples might correlate about .63 for total useof subordination, but below .59 forany meas-ure of a particular subordinate structure. The1000-word sample may, then, also be inade-quate and too unreliable for measuring specificsubordinate structures.

For the synopsis scores, however, the cor-relations between samples were higher, andfor the odd/even pages split they approacheda useful level. The measures of clause length,T-unit length, and sentence length appearedto have reached a level of reliability (r > . 69)that would be sufficient to obtain reliablemeasures of a group's characteristic writingbehavior. For individuals, one might still de-sire the slightly higher reliability than mightbe expected from 1000-word samples.

Table 2

Correlations for 17 Pairs of Writing MeasuresTaken from Two 500-Word Samples

SyntacticalStructures

Frequencyin 1000Words

Odd/EvenPagesAll Ss

First-Half/Second-Half

All Ss

Odd /EvenOdd /Even2a.aesly___GroupsMale Female High Low

Ss Ss Ability Ability

Adjective Clauses 10.49 .42 .35 .53 .31 .34 .5 2

Adverb Clauses 17.73 .41 .31 .46 .36 .5 2 .27Noun Clauses 14.21 .26 .08 .24 .27 .45 .02Clauses of Comparison .93 -.01 -.12 .05 -.07 .01 -.03Deferred Subject .16 .01 .03 .25 -.06 -.08 .20Cleft Sentence .20 .29 .11 .46 .08 .23 .35"The More, the Merrier" .03 -.02 -.01 -.03 -- --Subjunctive .01 -- -.01 -- -- -- --Special Which Clauses .19 .08 -.05 -.09 .30 -.09 .22Adjective Complement .24 -.04 .10 -.05 -.07 -.04 -.07Adverb Complement .13 .04 . 07 .19 -.06 .13 - .06

All Subordinate Clauses 44.33 .46 .39 .49 .43 .51 .39

Synopsis Scores Mean

Clause Length , 8.32 .69 .48 . 72 .64 . 72 .64Clauses per T-unit 1.59 .54 .22 .56 .5 2 .65 .42T-unit Length 13.22 .74 .56 .81 .68 .82 .61T-units per Sentence 1.25 .65 .48 .72 .51 .69 . 62

Sentence Length 16.47 .77 .62 .76 .79 .79 .75

5

In comparing the two methods of samplingfrom a writing sample, odd/even pages vs.first-half/second-half, it is obvious that theodd/even pages procedure was superior. Forthe synopsis scores the odd/even correlationsranged from .54 to .77 compared to a range of

/ to .62 for the first-half/second-half, Thatis sampling every other page represents theentire corpus of Writing more adequately thanif one considers only the first half. Perhapsthe ideal procedure would be to consider everyother sentence, or every third sentence, orevery fourth sentence, etc., depending on theproportion of the total corpus one wanted tosample.

In Table 2 the correlations between writingsamples taken from the odd/even split are pre-sented for male, female, high IQ, and low IQSs separately. Several differences betweenpairs of correlations are apparent, but whatthese differences indicate about writing per-formance is uncertain. Possibly the writingof the male and high IQ Ss was more stableand consistent since the correlations for thesetwo groups were generally higher than for fe-male and low IQ Ss. Consistency does not,however, imply maturity.

In Table 3 the intercorrelations between theodd/even pages synopsis scores are presented.All correlations greater than .26 in absolutevalue show significant relationship (a< .01)for 133 degrees of freedom. Note that measuresof clause length, clauses per T-unit, and T-units per sentence are largely independent

Table 3

measures of writing behavior. That is, a per-son who writes long clauses does not neces-sarily write more or less clauses than someonewho writes short clauses. If clauses per T-unit is considered a measure of subordination,and T-units per sentence a measure of coor-dination, it is apparent from Table 3 that atEighth Grade these two skills are uncorrelated.Clause length, sentence length, and T-unitlength, however, are significanay correlatedsince sentences are made up of T-units andclauses, and T-units include any clausespresent.

The following conclusions become apparentfrom the comparison of two 500-word samples:

1. Several structures (clauses of compari-son, deferred subject, cleft sentence,"the more, the merrier," subjunctive,special which clauses, adjective com-plement, and adverb complement) werevery infrequent in 500 words and also1000 words. Much larger samples ofwriting covering various writing situa-tions would be needed to make reliableestimates of their incidence.

2. Adjective, adverb, and noun clausestogether occurred about 5, 9, and 7times, respectively, in an average 500 -word sample of Eighth-Grade writing.The frequencies correlated from .26 to.4 2 from sample to sample (odd/evenpages). These correlations seem toolow for most purposes, and even the

Intercorrelations of the Five Synopsis Scoresin the Odd/Even Comparison*

Variable

Odd PagesClause LengthClauses per T-unitT-unit LengthT-units per SentenceSentence Length

Even PagesClause LengthClauses per T-unitT-unit LengthT-units per SentenceSentence Length

Odd Pagesw/C C/T W/T

(w /C)(C/T)(W /T)(T /S)(w/s)

(W /C)(C /T)

(W /T)(T /S).(w/s)

1.00 -.031.00

.77

.601.00

Even PagesT/S w/s w/C C/T W/T T/S w/S

-.22 .53-.05 .51-.21 .741.00 .49

1.00

. 69 .12 .58 -.28 .32

. 09 .54 .45 .03 .44.59 .45 .74 -.20 .54

-.01 -.06 .04 .65 .42. 53 .34 .63 .27 .77

1.00 -.051.00

.67

.701.00

-.15-.24-.281.00

.52

.47.72.45

1.00

* r > .26 is significant at the .01 level

6

higher estimates from a larger sample(1000-words might correlate .41 to .59)would be inadequate for measures ofindividual performance.

3. The total number of all subordinatestructures reached a sample-to-samplereliability of .46, also too unreliablefor most purposes.

4. The synopsis scores showed relativelyhigh levels of reliability compared tothe subordinate structure frequencies.

5. Though groups of Ss did not differ sig-nificantly from sample to sample, dif-ferent sampling procedures varied inhow well they represented a larger sam-ple. In the present case, comparingodd to even pages resulted in highercorrelations than comparing first-halfto second-half samples.

Thus, it appears that at least three of thesynopsis scores (clause length, clauses per

GPO 914-474-3

T-unit, and T-units per sentence) are non-redundant measures that may be obtained ob-jectively and reliably from a writing sample of500 words. Two other synopsis scores, sen-tence length and T-unit length, are also reli-able when obtained from 500-word samples,but reflect factors which are already measuredby the other three synopsis scores. Sentencelength and T-unit length are therefore corre-lated more highly with the other synopsisscores.

The reliability of the synopsis scores from500-word samples appears to compare favor-ably to the levels usually obtained in themeasurement of writing. Apparently, samplingsystematically (every other page or sentence)will provide a reliable representation of thelarger corpus of writing, as far as the synopsisscores are concerned. For measures of thefrequency of specific types of clauses, theonly option. appears to be the tabulation ofvery large writing samples over many types oftopics and situations.

7

REFERENCES

Anderson, J. E. An evaluation of various in-dices of linguistic development. ChildDevelopment, 1937, 8, 62-68.

Blount, N. S., Fredrick, W. C., & Johnson,S. L. The effect of a study of grammar onthe writing of eighth-grade students. Tech-nical Report from the Wisconsin Researchand Development Center for Cognitive Learn-

University of Wisconsin, 1968, No. 69.Chat los, J. W. Studies in language behavior.

IV. A statistical and comparative analysis

GPO 014-474-2

of individual written language samples.Psychological Review (Psychological Mono-graphs), 1944, 56, 77-111.

Frogner, E. Problems of sentence structure inpupils' themes. English Journal, 1933, 22,742-749.

Hunt, K. W. Grammatical structures writtenat three grade levels. National Council ofTeachers of English Research Report No. 3.Champaign, Illinois: National Council ofTeachers of English, 1965.

9